Based On Neural Network Medium Plate Rolling Force Prediction

The rolling process mathematical model is the core of the control system ofthe mill; the rolling force prediction model has an important impact on productthickness and plate type. With the market conditions increasing competitive,how to improve the product quality will be a very important issue, and theimportance of the calculation accuracy of the rolling process mathematicalmodel is more and more evident. The modern rolling control system has beenthat the hardware core is the computer as the software core is the mathematical,which the calculation accuracy of rolling process mathematical model is moreand more obvious. Rolling deformation process is very complex; there are manyvariables which have non-linear mapping relationship in the rolling process. Thetraditional calculation of rolling force model is derived under the conditions thatrolling material is uniformity, the plastic material is ideal and has nobroadening， deformation of the upper and lower is symmetry. So the traditionalmodel is difficult to accurately describe the change of the rolling process. weneed to improve the model or combination with other methods to improve theforecast level of rolling force model.Based on the hot rolling force mathematical model of plate, and in-depthanalysis of the characteristics of the Sims rolling force model in hot rolling fromthe Orowan deformation theory at the first, the main factors to determine theimpact of rolling force calculation accuracy are metal deformation resistance Kand stress state coefficientQ_P. For rolling force online forecast characteristic thatrequirements calculated fast, analytical solution is solved which consider rollerflattening factors. Then, knowledge of the BP neural network is introduced indetail, on the basis of the mathematical model of rolling force, a new rollingforce prediction model based on neural network is provided, the number of inputand output variables of the neural network model is reduced, using neuralnetwork correct product of K andQ_P, then the corrected values are substitutedinto the Sims model, Neural network which has the struct of5-20-1is offlinetraining by the actual production data in a hot-rolled steel four-high reversing mill. The results show that the proposed new rolling force prediction modelunder the premise of ensuring the rolling force prediction accuracy, andeffectively improve the convergence rate of the neural network. Finally, thescheme of rolling force prediction model of online applications is given, anddevelops the online application software which rolling force prediction based onneural network approach.The rolling force prediction model which rose in the thesis simplified theneural network correction module, so it is more suitable to the online application.Through the offline simulation, the feasibility of forecast is validated. In order tothat the application software applied in the actual production process, we need todo further optimization and improvement for developed prediction modelsoftware.